TD Bank

TD Bank faced mounting insurance claims costs driven by fraud, inefficient vendor relationships, and slow manual processes that struggled to keep pace with the scale of a major North American retail bank. Legacy transaction monitoring systems lacked the sophistication to accurately assess financial crime risk, resulting in both missed fraud and costly false positives. Claims resolution timelines were extended by manual workflows, while financial planning processes consumed significant staff time. The cumulative effect was a material and growing cost burden that demanded a structural response rather than incremental fixes.

TD deployed a multi-layered AI program targeting insurance claims costs through three coordinated workstreams: ML-enhanced transaction monitoring, vendor optimization, and end-to-end process reengineering. Machine learning models were integrated directly into the bank's existing transaction monitoring system, with additional models scheduled for rollout in subsequent quarters. In parallel, a data-driven financial crime risk evaluation methodology replaced earlier heuristic approaches, enabling more granular and accurate assessment of financial crime exposure. A generative AI Knowledge Management System was deployed first in contact centers, then scaled across more than 1,000 Canadian branches — illustrating the bank's core "build once, use many times" deployment principle, which prioritizes repeatable patterns to accelerate rollout and reduce delivery cost.

TD's AI investments generated $170 million in total value in 2025, validating the business case for continued scaled deployment. Specific outcomes from the insurance and fraud program include:

• $150M insurance cost reduction targeted in the medium term through fraud detection, vendor optimization, and process reengineering
• 50% reduction in time required to create financial plans
• Faster fraud detection and improved claims resolution speed and accuracy
• Contact center hold times reduced; branch staff now answer complex questions in seconds rather than navigating multiple systems

The bank's enterprise AI value target stands at $1 billion annually, with agentic AI projects — including real estate secured lending pre-adjudication — currently in scaling phases.

• Pairing AI with vendor and process reform compounds savings — TD's $150M target reflects all three levers working together, not AI in isolation.
• A 'build once, use many times' architecture significantly reduces deployment costs — the Knowledge Management System scaled from contact centers to 1,000+ branches without re-building from scratch.
• ML integration into existing transaction monitoring systems is lower-risk than replacement — TD added models incrementally to live infrastructure, maintaining continuity while improving coverage.
• Quantifying AI value explicitly (e.g., $170M in 2025) creates accountability and sustains investment — banks should instrument AI programs to produce auditable financial outcomes, not just operational metrics.